bzr-fastimport fails to incrementally re-import bzr-migration-docs from git

[Miklos Vajna]

Here is a script to reproduce the issue I have:

http://frugalware.org/~vmiklos/files/bzr-fastimport-bug-sh

Here is the output of the script on my machine:

http://frugalware.org/~vmiklos/files/bzr-fastimport-bug-out

This is r253 from trunk and git version 1.6.5.

Let me know if you can't reproduce the issue and you need more info.

Thanks.

[Yaroslav Halchenko (yarikoptic)]

Was about to file a new bug report but finally found at least the one describing the same symptoms ;)

Preamble: I was using git-bzr script (here is my clone of its Ruby version http://github.com/yarikoptic/git-bzr) in the beginning of last year and it was working fine... Recently I've tried to incrementally upgrade few of git-bzr'ed repositories and updates failed with similar 'KeyError'...
Decided to give a try to Python version of the same script from Matthew Brett (http://github.com/matthew-brett/git-bzr/) where he also describes that push back into bzr didn't work -- decided to look finally WTF is happening and it immediately gave me a clue:

Ugly face of the bug is revealed with a dump of marks file within bzr:
{{{
$> cat test-git/.git/bzr-git/bzr-repo-bzr-map
format=1
tmp.0
1 <email address hidden>
}}}

Whenever KeyError would be: KeyError: ':1'

More interesting that upon subsequent fastimport (after you fix marks file and add that leading colon) it would add a new mark WITH colon at the beginning:

{{{
$> cat test-git/.git/bzr-git/bzr-repo-bzr-map
format=1
tmp.0
:1 <email address hidden>
:3 <email address hidden>
}}}

So, the problem is buried in handling marks during export, and actual history of modifications of that file is quite interesting (sorry for non-bzrish format, I know only git):

{{{
commit 682805c7a9890f09587be709e94fe3188a9f65f5
Author: Ian Clatworthy <email address hidden>
Date: Sun Aug 23 18:35:41 2009 +1000

    backout git-bzr fix as it was breaking fast-export

diff --git a/bzr_exporter.py b/bzr_exporter.py
index 6ce20a0..a1a46d1 100755
--- a/bzr_exporter.py
+++ b/bzr_exporter.py
@@ -244,7 +244,7 @@ class BzrFastExporter(object):
         # Print the commit
         git_ref = 'refs/heads/%s' % (git_branch,)
         mark = ncommits + 1
- self.revid_to_mark[revid] = ':' + str(mark)
+ self.revid_to_mark[revid] = mark
         file_cmds = self._get_filecommands(parent, revid)
         self.print_cmd(self._get_commit_command(git_ref, mark, revobj,
             file_cmds))

commit b554c8ab84b0fb2b2d624b8a73120a4236abd49b
Author: Gonéri Le Bouder <email address hidden>
Date: Fri Aug 21 16:47:35 2009 +0200

    really put ':' in the commit mark

diff --git a/bzr_exporter.py b/bzr_exporter.py
index 2685731..6ce20a0 100755
--- a/bzr_exporter.py
+++ b/bzr_exporter.py
@@ -244,7 +244,7 @@ class BzrFastExporter(object):
         # Print the commit
         git_ref = 'refs/heads/%s' % (git_branch,)
         mark = ncommits + 1
- self.revid_to_mark[revid] = str(mark)
+ self.revid_to_mark[revid] = ':' + str(mark)
         file_cmds = self._get_filecommands(parent, revid)
         self.print_cmd(self._get_commit_command(git_ref, mark, revobj,
             file_cmds))

commit 6596f1441ade6718094708434c40a47a87d60e55
Author: Gonéri Le Bouder <email address hidden>
Date: Wed Aug 19 14:11:40 2009 +0200

    add the missing ":" since revid_to_mark are "committish"

diff --git a/bzr_exporter.py b/bzr_exporter.py
index d7f3435..2685731 100755
--- a/bzr_exporter.py
+++ b/bzr_exporter.py
@@ -243,7 +243,8 @@ class B...